What Drives AI Adoption in the Real World?

As artificial intelligence (AI) rapidly transforms the modern economy, public debate has focused heavily on how readily a job could be performed by AI—in other words, how much AI “exposure” the job has. The underlying assumption is that knowing which tasks AI is able to automate or augment will reveal how the technology is likely to reshape the labor market.

Yet there is a wide gap between what AI can do and whether workers actually use AI on the job. A new study by Ilse Lindenlaub, Ryungha Oh, María Alejandra Rodríguez, and Laura Veldkamp argues that exposure captures AI’s absolute advantage— its productivity in performing a task —whereas adoption depends on comparative advantage. Put simply, the relevant comparison is how much output AI delivers per dollar of user cost versus how much output a worker delivers per dollar of pay. To test this idea, the researchers combine two ingredients: a unique, nationally representative survey of actual AI use by German workers, linked to official worker and establishment records, including wages; and an economic model in which adoption is driven by comparative advantage. Together the data and model allow them to move beyond technical capability of AI by estimating AI user costs as well as worker productivity relative to pay. They then use these estimates to predict AI adoption and compare those predictions with actual workplace use. Their central finding is that technical capability—and thus AI exposure—alone is a poor predictor of whether an individual worker adopts AI.

A task might be technically feasible for AI, but if the user costs— such as operating expenses, workflow integration, output verification, regulatory compliance, and data privacy requirements—are too high, then adoption stalls. AI is also less likely to be adopted when a worker is exceptionally productive relative to their pay in a given task. Adoption is highest where AI combines strong capabilities with low user costs relative to the cost-effectiveness of human labor.

“Our forecasting exercise illustrates why the distinction between AI productivity and user costs matters for predicting AI diffusion. Projected adoption growth is not driven by improvements in AI productivity alone, nor by falling user costs alone. Rather, the largest adoption increases arise when AI becomes both more productive and cheaper or easier to use.”

This economic distinction dramatically shifts our understanding of AI's actual impact. Some occupations, like accountants, face massive technical exposure to AI, meaning that AI could automate many of their job tasks. And yet, steep costs of verifying the AI's work as well as privacy concerns, combined with relatively high levels of productivity among accountants, temper the technology's true comparative advantage. Conversely, primary school teachers have core tasks with lower AI exposure but also lower user costs—giving AI a compelling relative edge and prompting surprisingly robust adoption.

Looking ahead, the authors project that AI adoption will nearly double over the next three years, reaching about 80%. That would mean most workers use AI for at least one task in their occupation. Yet wider adoption does not necessarily mean broader use: AI is projected to spread to more workers without being used across many more tasks per worker, leaving the breadth of use largely unchanged. Crucially, little of the predicted increase in adoption is due to the predicted increase of AI’s technical capability. Instead, falling costs are the primary driver. Predicting the future of work therefore requires looking beyond what AI can do in principle to whether it is profitable to use in practice. What ultimately matters is the economic trade-off between human skill and machine efficiency.